Features
Seven capabilities. One verified record underneath all of them.
Everything below writes to the same append-only, SHA-256 chained audit trail, so your monitoring, your enforcement, and your compliance evidence can never disagree.
Realtime agent monitoring
See every AI agent call as it happens.
Each completion, tool call, and decision your agents make becomes a structured event on a live dashboard, classified in realtime against the MAST failure taxonomy from UC Berkeley research presented at NeurIPS 2025.
When something goes wrong, you are not reconstructing the past from application logs. You are reading a verified record of what actually happened.
For developers and security teams
Policy engine
Define what your agents can and cannot do.
Write rules in plain terms: which tools an agent may call, which data it may touch, which actions need a human. Each incoming action is evaluated against your policies and resolved to allow, block, or require review.
Policies live alongside the audit trail, so every enforcement decision is recorded with the rule that triggered it.
For security teams and compliance managers
Human-in-the-Loop
Route sensitive decisions to a human reviewer.
High-risk actions pause in a review queue until a named reviewer approves, rejects, or escalates. The agent waits; the human decides; the decision is logged with reviewer identity and timestamp.
This is the workflow EU AI Act Article 14 and GDPR Article 22 expect you to have, shipped as a product feature rather than a policy document.
For compliance managers and legal teams
Compliance reporting
Generate audit-ready reports for any framework.
Pick GDPR, HIPAA, EU AI Act, SOX, MiFID II, NIST AI RMF, PIPEDA, or AIDA and a date range. MastGuard builds the report directly from your chain-verified audit data and exports a PDF formatted the way a reviewer expects.
Evidence packages go further: a ZIP with the verified audit log, the narrative report, and a manifest of file hashes. Enterprise adds up to 7-year immutable retention.
For compliance, legal, and the board
RedScan adversarial testing
Test your AI agents before attackers do.
RedScan runs 468 adversarial test cases against your agent: prompt injection, jailbreaks, scope escape, and multi-step chains. You get a pass or fail per attack category and an AI Risk Score from 0 to 100.
The output is a board-ready PDF mapped to NIST AI RMF, ISO 42001, and the EU AI Act. Find the open path in a test run, not in an incident report.
For security teams and developers
MCP Monitor
Full visibility into every MCP tool call.
The MCP proxy sits between your agent and its Model Context Protocol servers. Every tools/call is checked against your tool scope, logged with latency and an output hash, and blocked when it falls outside policy.
The MCP Monitor dashboard shows per-tool frequency, blocked calls, and per-session timelines, so the newest part of your agent stack is no longer the least observable.
For developers and security architects
AI-BOM and ProvenanceGuard
Know exactly what models and data power your agents.
ProvenanceGuard screens training datasets for poisoning using absolute-count cluster analysis: research shows as few as 250 corrupted documents can backdoor a model at almost any scale, which percentage thresholds will miss.
Every dataset gets an AI Bill of Materials aligned to EU AI Act Article 11, so when a regulator asks what your model learned from, you have the record.
For compliance managers and procurement
See it on your own agents.
The free tier monitors a production agent end to end: 50,000 events a month, realtime detection, and a 7-day audit trail.